Hybrid Deep Learning Architectures Using Markovian Sampling Support Vector Machine for Multimodal Breast Cancer Classification
V. Mahalakshmi, G. Sasikala · IETE Journal of Research · 2025
Breast cancer still remains a prevalent disease that causes high mortality in women across the world and requires trustworthy, effective, and precise computer-aided detection methods. A novel classification pipeline that uses a fused bilateral guided median (BGM) filter to denoise multimodal breast cancer images by preserving primary tissue structures and improving image feature extraction has been developed in this context to improve breast cancer diagnosis. Further, it also combines deep features that extracted using hybrid 7-layer convolutional neural networks (CNN), ResNet-50, VGG-19, and DenseNet-121 with an advanced Markovian sample support vector machine (MSSVM) classifier that aims to overwhelm the conventional machine learning (ML) classifiers like support vector machines (SVM), naïve Bayes (NB), decision trees (DT), and random forests (RF). Compared to traditional convolutional neural network architectures, the proposed framework improves the classification accuracy by an average of 97.79% (mammogram images—D-I), 90.52% (ultrasound images—D-II), and 82.78% (histopathology images—D-III) and reduces the average training time to about 148.56 s. Notably, the DenseNet-121 paired with Markovian sample SVM outperformed the other architectures in terms of capturing deep hierarchical features and handling hard decisions. These findings emphasize the effective strategy of hybrid architectures towards the development of robust and clinically pertinent diagnosis systems for breast cancer in real time.